OpenAI Launches GPT-5.5-Cyber for Security Researchers

OpenAI has released GPT-5.5-Cyber, a cybersecurity-focused version of its latest model, available in limited preview through its Trusted Access for Cyber program. The model scored 81.9% on the CyberGym benchmark and can generate vulnerability exploitation plans and validate them through simulated cyberattacks, making it useful for automated red teaming. Access is initially restricted to defenders securing critical infrastructure, with stronger verification and enhanced misuse monitoring in place. The model competes with Anthropic's Claude Mythos Preview, which is similarly positioned for advanced cybersecurity research.